Automated Waste Sorting

Image classification and object detection for automated waste sorting.

This project was developed as part of an international collaborative initiative between the Lancaster University campuses of Leipzig (Germany) and Bailrigg (England).

HW: KUKA industrial robots, the same ones used in car assembly lines, capable of handling loads of 180kg and moving at 3m/s.

Motivation: By innovating in waste management, we hope to contribute toward sustainable solutions for one of the world’s growing environmental concerns. Our long-term vision is to extend this technology to more critical areas, such as nuclear waste management, where safe and efficient handling is of utter importance.

I led the computer vision team, focusing on object detection and image classification. For the code, see repo.

Dataset

Custom curated dataset of labeled trash/waste images. These are spread across 6 subfolders (waste types). The dataset can be downloaded here.

Model

Fine-tuned pre-trained ResNet50 architecture. Model weights can be downloaded here.

Demo

Demo: Fully autonomous arm classifying Pepsi can as "metal" and sorting it to its appropriate bin.

Credits

  • Matias Barandiaran
  • Parichay Sachdev
  • Mustafa Azizi
  • Athar Syed
  • Osvaldo Catine
  • Mikelis Kamepe
  • Inderjot Sitt
  • Isaac Richardson
  • Andre Mariucci
  • Toby Vermon